11.4 Audio and speech machine learning
You can build a keyword spotter or a transcription pipeline.
Before:06. Deep Learning
Audio machine learning spans classification, keyword spotting, transcription with Whisper, diarization and synthesis — and a signal-processing background removes the front-end mystery entirely. It sits at the applied centre of the bridge module. The deployment gap is the lesson: pipelines tested only on clean audio meet noise, accents, far-field microphones and codec artefacts in the field, and the gap between lab and deployment is where audio products die.
Work through these
Audio classification and event detection
Classifying whole audio clips and spotting specific events within them. This is the entry point to audio machine learning and it reuses the image techniques from the vision module.
ASR: Whisper and streaming transcription
Turning speech into text, including the case where it must be done as the person speaks. The streaming case is substantially harder than the batch one.
Speaker identification and diarization
Recognising who is speaking, and separating a recording into who spoke when. Both are needed for any system that handles meetings or calls.
TTS and voice interfaces
Producing speech from text, and the loop of listening and speaking that makes a voice interface. Latency is what decides whether the result feels like a conversation.
Sign in to keep your progress.
Free resources
We haven't checked most of these for screen reader use yet.
Links last checked 29 Aug 2026.
Stuck here?
Ask a mentor. A real person answers, and they can see exactly which topic you're on. Usually within a couple of working days.
Checking your session…
Topics shown in module order.